Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Typhoon OCR: Open Vision-Language Model For Thai Document Extraction

Surapon Nonesung, Natapong Nitarach, Teetouch Jaknamon, Pittawat Taveekitworachai, Kunat Pipatanakul

TL;DR

Typhoon OCR tackles Thai document extraction by fine-tuning an open vision-language backbone on a Thai-focused corpus, using a two-mode supervision and a multi-stage data pipeline to handle diverse layouts. It demonstrates that compact models (e.g., 2B parameters) can match or exceed larger proprietary systems on structured Thai documents while delivering lower inference costs. The work emphasizes the value of open, adaptable multimodal models for low-resource languages and real-world documents, supported by synthetic data and Thai-specific VQA augmentation. Limitations include degraded performance on severely degraded inputs and the need for broader language coverage and higher-level reasoning tasks in future work.

Abstract

Document extraction is a core component of digital workflows, yet existing vision-language models (VLMs) predominantly favor high-resource languages. Thai presents additional challenges due to script complexity from non-latin letters, the absence of explicit word boundaries, and the prevalence of highly unstructured real-world documents, limiting the effectiveness of current open-source models. This paper presents Typhoon OCR, an open VLM for document extraction tailored for Thai and English. The model is fine-tuned from vision-language backbones using a Thai-focused training dataset. The dataset is developed using a multi-stage data construction pipeline that combines traditional OCR, VLM-based restructuring, and curated synthetic data. Typhoon OCR is a unified framework capable of text transcription, layout reconstruction, and document-level structural consistency. The latest iteration of our model, Typhoon OCR V1.5, is a compact and inference-efficient model designed to reduce reliance on metadata and simplify deployment. Comprehensive evaluations across diverse Thai document categories, including financial reports, government forms, books, infographics, and handwritten documents, show that Typhoon OCR achieves performance comparable to or exceeding larger frontier proprietary models, despite substantially lower computational cost. The results demonstrate that open vision-language OCR models can achieve accurate text extraction and layout reconstruction for Thai documents, reaching performance comparable to proprietary systems while remaining lightweight and deployable.

Typhoon OCR: Open Vision-Language Model For Thai Document Extraction

TL;DR

Typhoon OCR tackles Thai document extraction by fine-tuning an open vision-language backbone on a Thai-focused corpus, using a two-mode supervision and a multi-stage data pipeline to handle diverse layouts. It demonstrates that compact models (e.g., 2B parameters) can match or exceed larger proprietary systems on structured Thai documents while delivering lower inference costs. The work emphasizes the value of open, adaptable multimodal models for low-resource languages and real-world documents, supported by synthetic data and Thai-specific VQA augmentation. Limitations include degraded performance on severely degraded inputs and the need for broader language coverage and higher-level reasoning tasks in future work.

Abstract

Document extraction is a core component of digital workflows, yet existing vision-language models (VLMs) predominantly favor high-resource languages. Thai presents additional challenges due to script complexity from non-latin letters, the absence of explicit word boundaries, and the prevalence of highly unstructured real-world documents, limiting the effectiveness of current open-source models. This paper presents Typhoon OCR, an open VLM for document extraction tailored for Thai and English. The model is fine-tuned from vision-language backbones using a Thai-focused training dataset. The dataset is developed using a multi-stage data construction pipeline that combines traditional OCR, VLM-based restructuring, and curated synthetic data. Typhoon OCR is a unified framework capable of text transcription, layout reconstruction, and document-level structural consistency. The latest iteration of our model, Typhoon OCR V1.5, is a compact and inference-efficient model designed to reduce reliance on metadata and simplify deployment. Comprehensive evaluations across diverse Thai document categories, including financial reports, government forms, books, infographics, and handwritten documents, show that Typhoon OCR achieves performance comparable to or exceeding larger frontier proprietary models, despite substantially lower computational cost. The results demonstrate that open vision-language OCR models can achieve accurate text extraction and layout reconstruction for Thai documents, reaching performance comparable to proprietary systems while remaining lightweight and deployable.
Paper Structure (33 sections, 5 figures, 5 tables)

This paper contains 33 sections, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Typhoon OCR
  3. Data
  4. Dataset Creation Pipeline
  5. Stage 1
  6. Stage 2
  7. Stage 3
  8. Stage 4
  9. Dataset Statistics
  10. Experimental Setup
  11. Training
  12. Evaluation
  13. Metrics
  14. Benchmark
  15. Protocol
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Overview of Typhoon OCR, illustrating supported input document types and the corresponding structured output representations.
  • Figure 2: Overview of the multi-stage dataset construction pipeline used to generate training data for Typhoon OCR under Structure Mode.
  • Figure 3: Composition of the training corpus used for Typhoon OCR. Figure \ref{['fig:dataset_dis']} shows the allocation of samples between Default Mode and Structure Mode supervision, while Figure \ref{['fig:dataset_by']} presents the relative contributions of different data sources.
  • Figure 4: Multi-stage pipeline for generating synthetic Thai document images for OCR training.
  • Figure 5: Training dataset distribution for Typhoon OCR V1.5